CN105391755B

CN105391755B - Data processing method, apparatus and system in a kind of distributed system

Info

Publication number: CN105391755B
Application number: CN201510644448.1A
Authority: CN
Inventors: 薛贞文
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2015-09-30
Filing date: 2015-09-30
Publication date: 2018-10-19
Anticipated expiration: 2035-09-30
Also published as: CN105391755A

Abstract

The invention discloses data processing method, apparatus and systems in a kind of distributed system, for solving the problem of to complete the data-handling efficiency of distributed system could be caused to reduce based on the progress data processing of new subregion view since calculate node needs to wait for the task based on former subregion view.The data processing method includes：When the first calculate node updates the data using first according to the first subregion view and is updated operation to first data, the second calculate node receives the second subregion view；Second calculate node updates the data using second according to the second subregion view and is updated operation to first data；Wherein, the first moment of the first subregion view instruction stores at least two memory nodes of first data, and at least two memory nodes that first data are stored with the second moment of the second subregion view instruction are not exactly the same.

Description

Data processing method, apparatus and system in a kind of distributed system

Technical field

Data processing method, device and it is the present invention relates to field of data storage, more particularly in a kind of distributed system System.

Background technology

Distributed system refers to being distributed to computing resource and storage resource on different nodes, by network by each node The system connected.Distributed system can be accomplished a task by managing and controlling multiple nodes, but externally performance For a complete system of independence, user's perception is less than distributed system task in internal execution details.

Fig. 1 is the schematic diagram of distributed system, wherein client for receiving user demand, (ask by such as write request and reading Ask), calculate node is responsible for distribution routing, and memory node is responsible for single-deck management and actual storage, in addition, further including pipe in system Node is managed, is responsible for being managed multiple calculate nodes and memory node.

Distributed system preserves data using more replication policies, i.e., same data is stored on multiple memory nodes, with Avoid loss of data caused by memory node failure, the reliability of enhancing data storage.Meanwhile system will also ensure same data Multiple copies consistency, i.e., with multiple memory nodes store same data when, it is ensured that stored on multiple memory nodes Data are consistent.

Chain type replicanism is a kind of strategy ensureing more copy consistencies, and reference Fig. 2 stores multiple storage sections of data There are chain relationships between point, are known as head node (and can be described as host node), the memory node of last-of-chain positioned at the memory node of chain head Referred to as tail node, calculate node carries out data write operation when carrying out write operation first in head node, then in chain type knot Write operation is carried out in node 2 in structure after head node, and so on, until be written data in tail node, calculate node is to visitor Family end returned data is written successfully and when reading data, and calculate node can only read the data stored in tail node.

In a distributed system, each calculate node is preserved a by the unified management of management node and updating maintenance Mapping table, referred to as subregion view preserves the information for the multiple memory nodes for storing each data in subregion view, calculates Node determines the memory node of each data with specific reference to the subregion view of preservation, and then asks corresponding memory node execution pair It should operate.

In actual conditions, storing the memory node of data may change, node of the management node in storage data New subregion view can be generated based on the memory node for storing data after change after change, and new subregion view is sent to Each calculate node.If there is calculate node is when receiving new subregion view, having the task based on former subregion view, (such as data are more New operation) not yet to complete, then remaining calculate node needs first to wait for the calculate node that will carry out based on the task of former subregion view It finishes, new subregion view could be based on and carry out data manipulation.Calculate node due to receiving new subregion view cannot be based on immediately New subregion view carries out data manipulation, will lead to task blocking, reduces the data-handling efficiency of distributed system.

Invention content

The embodiment of the present invention provides data processing method, apparatus and system in a kind of distributed system, for solve due to Calculate node needs to wait for the completion of the task based on former subregion view, and could be based on new subregion view progress data processing causes to be distributed The problem of data-handling efficiency of formula system reduces.

In a first aspect, data processing method in a kind of distributed system of offer of the embodiment of the present invention, the distributed system Including management node, the first calculate node, the second calculate node and multiple memory nodes, the management node, first meter It is communicated between operator node, second calculate node and the multiple memory node, the method includes：

It is updated the data using first according to the first subregion view in first calculate node and first data is carried out When update operation, second calculate node receives the second subregion view；

Second calculate node according to the second subregion view using second update the data to first data into Row update operation；

Wherein, the first subregion view and the second subregion view are generated by the management node；Described One subregion view is used to indicate at least two memory nodes for storing first data at the first moment；The second subregion view It is used to indicate at least two memory nodes that the second moment stored first data；Wherein, first moment and described the Two moment are different；First moment stores at least two memory nodes of first data and second moment stores institute At least two memory nodes for stating the first data are not exactly the same.

With reference to first aspect, in the first possible realization method of first aspect, the first subregion view instruction Storage first data at least two memory nodes and the second subregion view instruction storage first data At least two memory nodes in include the first memory node；

Second calculate node according to the second subregion view using second update the data to first data into Row update operation, including：

Second calculate node competes the write permission of first memory node with first calculate node；

Second calculate node updates the data after the write permission for obtaining first memory node using described second Operation is updated to first data in first memory node.

The possible realization method of with reference to first aspect the first, in second of possible realization method of first aspect In, first memory node is at least two memory nodes of storage first data of the first subregion view instruction In memory node in addition to primary storage node, and the storage institute that first memory node indicates for the second subregion view State the primary storage node at least two memory nodes of the first data；

Second calculate node competes the write permission of first memory node with first calculate node, including：

Second calculate node determines the primary storage node for storing first data according to the second subregion view For first memory node；

Second calculate node is updated operation to first data to first memory node transmission and asks It asks；

Wherein, main memory of first calculate node in storage first data indicated the first subregion view First data in storage node are updated after operation, and first calculate node is sent to first memory node First data are updated with the request of operation；First memory node will be updated operation to first data Write permission authorize the corresponding calculate node of request that operation is updated to first data first received.

With reference to first aspect, in the third possible realization method of first aspect, the second subregion view instruction At least two memory nodes of storage first data include the first memory node and the second memory node；

Second calculate node updates the data described first stored to first memory node using described second Data are updated, and form the data of the first version in first memory node；

Second calculate node determines that the corresponding version of first data that second memory node preserves is less than The previous version of the first version；

First data that second calculate node is preserved in second memory node are updated to described first After the data of the previous version of version, updated the data to the first edition in second memory node using described second The data of this previous version are updated, and form the data of the first version in second memory node.

The third possible realization method with reference to first aspect, in the 4th kind of possible realization method of first aspect In, second calculate node determines the corresponding version of first data that second memory node preserves less than described the The previous version of one version, including：

Second calculate node is updated to the second memory node transmission data asks, and the data update request is used First data are updated using the data that described second updates the data as first version in request；

Second calculate node receives the error information that second memory node returns, and the error information shows institute State the previous version that the corresponding version of first data in the second memory node is less than the first version；

First data that second calculate node is preserved in second memory node are updated to described first After the data of the previous version of version, updated the data to the first edition in second memory node using described second The data of this previous version are updated, including：

After second calculate node sets duration after receiving the error information, again to second memory node Send the data update request；

Second calculate node receives the data update success message that second memory node returns, wherein described Second memory node is updated to by first calculate node previous version of the first version in first data of storage After this data, the data update request that second calculate node is sent is responded, and in first data by more After newly being updated the data for described second the data update success message is returned to first calculate node.

Second aspect, the embodiment of the present invention provide data processing equipment in a kind of distributed system, the distributed system Including management node, the first calculate node, the data processing equipment and multiple memory nodes, the management node, described It is communicated between one calculate node, the data processing equipment and the multiple memory node, the data processing equipment includes：

Receiving module, for being updated the data to described using first according to the first subregion view in first calculate node When first data are updated operation, the second subregion view is received；

Update module carries out more first data for being updated the data using second according to the second subregion view New operation；

In conjunction with second aspect, in the first possible realization method of second aspect, the first subregion view instruction Storage first data at least two memory nodes and the second subregion view instruction storage first data At least two memory nodes in include the first memory node；

The update module, is specifically used for：

The write permission of first memory node is competed with first calculate node；

After the write permission for obtaining first memory node, updated the data to the first storage section using described second First data in point are updated operation.

In conjunction with the first possible realization method of second aspect, in second of possible realization method of second aspect In, first memory node is at least two memory nodes of storage first data of the first subregion view instruction In memory node in addition to primary storage node, and the storage institute that first memory node indicates for the second subregion view State the primary storage node at least two memory nodes of the first data；

The update module is specifically used for：

Determine that the primary storage node for storing first data is the first storage section according to the second subregion view Point；

The request that first data are updated with operation is sent to first memory node；

In conjunction with second aspect, in the third possible realization method of second aspect, the second subregion view instruction At least two memory nodes of storage first data include the first memory node and the second memory node；

The update module is specifically used for：

First data for storing first memory node are updated the data using described second to be updated, and are formed The data of first version in first memory node；

Determine the corresponding version of first data of the second memory node preservation less than before the first version One version；

The previous version of the first version is updated in first data that second memory node preserves After data, the number of the previous version to the first version in second memory node is updated the data using described second According to being updated, the data of the first version in second memory node are formed.

In conjunction with the third possible realization method of second aspect, in the 4th kind of possible realization method of second aspect In, the update module is specifically used for：

It updates and asks to the second memory node transmission data, the data update request is for asking with described second It updates the data and first data is updated as the data of first version；

The error information that second memory node returns is received, the error information shows in second memory node The corresponding version of first data be less than the first version previous version；

After receiving the error information after setting duration, the data update is sent to second memory node again Request；

Receive the data update success message that second memory node returns, wherein second memory node is being deposited After first data of storage are updated to the data of previous version of the first version by first calculate node, response The data update request that the data processing equipment is sent, and it is updated to the second update number in first data The data update success message is returned according to backward first calculate node.

The third aspect, the embodiment of the present invention provide calculate node in a kind of distributed system, are applied in distributed system, The distributed system further includes management node, the first calculate node and multiple memory nodes, the management node, described first It is communicated between calculate node, the calculate node and the multiple memory node, the calculate node includes：

Processor is communicated with input/output interface, for being used according to the first subregion view in first calculate node First updates the data when being updated operation to first data, indicates that the input/output interface receives the second subregion and regards Figure；

The input/output interface is used to receive the second subregion view according to processor instruction；

The input/output interface is additionally operable to reception second and updates the data；

The processor be additionally operable to according to the second subregion view using second update the data to first data into Row update operation；

In conjunction with the third aspect, in the first possible realization method of the third aspect, the first subregion view instruction Storage first data at least two memory nodes and the second subregion view instruction storage first data At least two memory nodes in include the first memory node；

The processor is updated the data using second according to the second subregion view and is updated to first data Operation, including：

In conjunction with the first possible realization method of the third aspect, in second of possible realization method of the third aspect In, first memory node is at least two memory nodes of storage first data of the first subregion view instruction In memory node in addition to primary storage node, and the storage institute that first memory node indicates for the second subregion view State the primary storage node at least two memory nodes of the first data；

The processor competes the write permission of first memory node with first calculate node, including：

Indicate that the input/output interface is updated operation to first memory node transmission to first data Request；

The input/output interface be additionally operable to first memory node send to first data carry out described in more The request newly operated；

In conjunction with the third aspect, in the third possible realization method of the third aspect, the second subregion view instruction At least two memory nodes of storage first data include the first memory node and the second memory node；

In conjunction with the third possible realization method of the third aspect, in the 4th kind of possible realization method of the third aspect In, the processor determines that the corresponding version of first data that second memory node preserves is less than the first version Previous version, including：

Indicate that the input/output interface receives the error information that second memory node returns, the error information table The corresponding version of first data in bright second memory node is less than the previous version of the first version；

The input/output interface is additionally operable to send the data update request to second memory node；Reception connects Receive the error information that second memory node returns；

First data that the processor is preserved in second memory node are updated to the first version After the data of previous version, updated the data to before the first version in second memory node using described second The data of one version are updated, including：

After receiving the error information after setting duration, indicate the input/output interface again to second storage Node sends the data update request；

Indicate that the input/output interface receives the data update success message that second memory node returns；

The input/output interface is additionally operable to send the data update request to second memory node again；It connects Receive the data update success message that second memory node returns；

Wherein, second memory node is updated in first data of storage by first calculate node described After the data of the previous version of first version, responds the data update that second calculate node is sent and ask, and First data be updated to described second update the data after to first calculate node return to data update success Message.

Fourth aspect, the embodiment of the present invention provide a kind of distributed system, and the distributed system includes management node, the One calculate node, the second calculate node and multiple memory nodes, the management node, first calculate node, described second It is communicated between calculate node and the multiple memory node；

The management node, for generate be used to indicate the first moment store first data at least two storage section First subregion view of point, and generate at least two memory nodes for being used to indicate that the second moment stored first data Second subregion view, wherein first moment is different from second moment；First moment stores first data At least two memory nodes stored with second moment first data at least two memory nodes it is not exactly the same；

First calculate node, for according to the first subregion view using first update the data to first data into Row update operation；

Second calculate node, for using the first update number according to the first subregion view in first calculate node When according to being updated operation to first data, the second subregion view is received, second is used according to the second subregion view It updates the data and operation is updated to first data.

In conjunction with fourth aspect, in the first possible realization method of fourth aspect, the first subregion view instruction Storage first data at least two memory nodes and the second subregion view instruction storage first data At least two memory nodes in include the first memory node；

Second calculate node is used for：It is updated the data to first number using second according to the second subregion view It is operated according to being updated, including：

In conjunction with fourth aspect, in second of possible realization method of fourth aspect, the second subregion view instruction At least two memory nodes of storage first data include the first memory node and the second memory node；

The one or more technical solutions provided in the embodiment of the present invention, have at least the following technical effects or advantages：

In the embodiment of the present invention, it is based on first the first data of subregion view pair in the first calculate node and carries out data update behaviour When being not finished, the second calculate node can be based on second the first data of subregion view pair and carry out data update operation.Due to not Terminated with the data update operation waited for based on the first subregion view, you can based on second point issued after the first subregion view Area view carries out data update operation, and the stand-by period and business processing for reducing business processing take, avoid because based on The obstruction of task based on the second subregion view caused by the task of first subregion view is delayed, improve distributed system into The efficiency of row data manipulation.

Description of the drawings

To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment Attached drawing is briefly introduced.

Fig. 1 is the schematic diagram of distributed system；

Fig. 2 is the schematic diagram of Linked Storage Structure；

Fig. 3 is the flow diagram of data processing method in the embodiment of the present invention；

Fig. 4 is another flow diagram of data processing method in the embodiment of the present invention；

Fig. 5 is the schematic diagram that memory node changes in the embodiment of the present invention；

Fig. 6 is the another flow diagram of data processing method in the embodiment of the present invention；

Fig. 7 is another schematic diagram that memory node changes in the embodiment of the present invention；

Fig. 8 is the structural schematic block diagram of data processing equipment in the embodiment of the present invention；

Fig. 9 is the structural schematic block diagram of calculate node in the embodiment of the present invention.

Specific implementation mode

Management node in distributed system is responsible for safeguarding the information of the memory node of data, generates the mark of data With the mapping table of the storage location of data, and the mapping table is sent to each calculate node, calculate node protects the mapping table It deposits on local storage, which is also known as the subregion view in calculate node, and subregion of the calculate node based on preservation regards Figure determines the memory node of data, and then initiates data service (such as read operation, write operation) to corresponding memory node and ask.

Data processing method, apparatus and system in a kind of distributed system of offer of the embodiment of the present invention, below by attached drawing And specific embodiment is described in detail technical solution of the present invention, it should be understood that in the embodiment of the present invention and embodiment Specific features are the detailed description to technical solution of the present invention, in the absence of conflict, the embodiment of the present invention and implementation Technical characteristic in example can be combined with each other.

Fig. 3 is the flow diagram of data processing method in distributed system, which includes the following steps：

Step 101：It is updated the data using first according to the first subregion view in the first calculate node and the first data is carried out When update operation, the second calculate node receives the second subregion view；

Step 102：Second calculate node is updated the data using second according to the second subregion view and is carried out more to the first data New operation；

Wherein, the first subregion view and the second subregion view are generated by management node；First subregion view is for referring to Show that the first moment stored at least two memory nodes of the first data；Second subregion view is used to indicate the second moment storage first At least two memory nodes of data；Wherein, the first moment is different from the second moment；First moment stored the first data at least At least two memory nodes that two memory nodes store the first data with the second moment are not exactly the same.

Specifically, distributed system includes management node, the first calculate node, the second calculate node and multiple storages section Point, wherein the first calculate node is two meters at least two calculate nodes in distributed system with the second calculate node Operator node.With continued reference to Fig. 1, pass through network between management node, the first calculate node, the second calculate node and multiple memory nodes It is communicated.

At the first moment, management node generates the first subregion view, and the first subregion view is used to indicate the storage of the first moment At least two memory nodes of the first data.After the first moment, storage of the management node because detecting the first data of storage Node changes, and generates the second subregion view at the second moment, the second subregion view is used to indicate the second moment storage first At least two memory nodes of data.Since the memory node that the second subregion view embodies is used to store first after changing The information of the memory node of data, so at least two storages of the first data of the first moment storage of the first subregion view instruction Node, at least two memory nodes that the first data are stored with the second moment of the second subregion view instruction are not exactly the same.

After management node generates the second subregion view, the second subregion view can be sent to each calculate node, so that Calculate node can be based on the second subregion view and carry out data manipulation (as read data manipulation, data writing operation).

In step 101, when the second calculate node receives the second subregion view, still there is the first calculate node according to first point Area's view updates the data using first and is updated operation to the first data.

In step 102, the second calculate node receives the request of client, executes the request of client, is calculated first In the case that node not yet terminates the update operation of the first data according to the first subregion view, updated the data pair using second First data are updated operation.

Due to without waiting for based on the first subregion view data update operation terminate, you can based on the first subregion view it The the second subregion view issued afterwards carries out data update operation, and the stand-by period and business processing for reducing business processing take, The obstruction because of the task based on the second subregion view caused by the task based on the first subregion view is delayed is avoided, is improved The efficiency of the progress data manipulation of distributed system.

Optionally, a kind of situation being likely to occur in the embodiment of the present invention is：The storage first of first subregion view instruction At least two memory nodes of data and at least two memory nodes of the first data of storage of the second subregion view instruction wrap Containing the first memory node.In this case, in order to avoid the first calculate node and the second calculate node are for the first storage The data manipulation of node mutually conflicts, and introduces competition mechanism in the embodiment of the present invention.

Step 102：Second calculate node is updated the data using second according to the second subregion view and is carried out more to the first data New operation, when it is implemented, with reference to Fig. 4, includes the following steps：

Step 1021：Second calculate node competes the write permission of the first memory node with the first calculate node；

Step 1022：Second calculate node updates the data pair after the write permission for obtaining the first memory node using second The first data in first memory node are updated operation.

Specifically, when carrying out data update operation to the first memory node, the write permission of the first memory node is obtained. In the present embodiment, when carrying out data update operation according to first the first data of subregion view pair due to the first calculate node, need It is updated the data using first and the first data in first memory node is updated, and data are carried out according to the second subregion view The second calculate node for updating operation also needs to carry out data update operation in the first memory node.

In the embodiment of the present invention, the second calculate node competes the write permission of the first memory node with the first calculate node, such as The second calculate node of fruit competes successfully, then the second calculate node is updated the data using second to the first number in the first memory node It is operated according to being updated., whereas if the first calculate node competes successfully, then the first calculate node first is updated the data to first The first data in memory node are updated operation.

In above-mentioned technical proposal, in the first calculate node being updated according to first the first data of subregion view pair, with And it is intended to first in the first memory node according to the second calculate node that second the first data of subregion view pair are updated When data are updated, by allowing the first calculate node to be avoided with the write permission of the second calculate node the first memory node of competition Concurrency conflict.

Optionally, in the embodiment of the present invention, the first calculate node is intended to the second calculate node in the first memory node A kind of possible situation for being updated of the first data be：First memory node is that the storage first of the first subregion view instruction counts According at least two memory nodes in memory node in addition to primary storage node, and the first memory node is the second subregion view Primary storage node at least two memory nodes of the first data of storage of instruction.

And step 1021：Second calculate node competes the write permission of the first memory node with the first calculate node, specific real Shi Shi may include steps of：

Second calculate node determines that the primary storage node of the first data of storage is that the first storage saves according to the second subregion view Point；

Second calculate node sends the request that the first data are updated with operation to the first memory node；

Wherein, the first calculate node in the primary storage node of the first data of storage indicated the first subregion view One data are updated after operation, and the first calculate node is updated operation to the transmission of the first memory node to the first data Request；First memory node the write permission that the first data are updated with operation is authorized first receive to the first data carry out Update the corresponding calculate node of request of operation.

Specifically, the second calculate node updated the data using second first data are updated when, first in basis The primary storage node that second subregion view is determined carries out data update in i.e. the first memory node, therefore, the second calculate node The request that the first data are updated with operation is sent to the first memory node.

And the first calculate node is regarded in the first subregion first when being updated according to first the first data of subregion view pair The first data are updated in the primary storage node that figure is determined, after update, are continued to true according to the first subregion view In (spare) memory node in addition to primary storage node at least two memory nodes of the first data of storage made First data are updated, wherein include being updated to the first data in the first memory node.

Therefore, the first memory node will receive two calculate nodes (the first calculate node and the second calculate node) hair The request being updated to the first data sent, the second calculate node will compete writing for the first memory node with the first calculate node Permission.Wherein, competition refers to sending to same data to same memory node in two or more calculate nodes When data update is asked, only one of which calculate node can obtain write permission, and the calculate node for obtaining write permission can be at Work(carries out data update, and the data update request for competing the calculate node of failure will not be responded by memory node.Specific implementation When, the second calculate node competes the write permission of the first memory node, including following realization method with the first calculate node：

First, the first calculate node and the second calculate node are to the first memory node to the first memory node transmission data Update request, the first memory node determine authorize write permission to which calculate node according to the sequencing for receiving request, that is, If the first memory node first receives the data update request of the first calculate node transmission, write permission is authorized to the first calculating Node then authorizes write permission to the second calculating section, whereas if first receiving the data update request of the second calculate node transmission Point.

Second, the first calculate node and the second calculate node are to the first memory node to the first memory node transmission data Update request, the first memory node determine authorize write permission to which according to the sequencing of calculate node transmission data update request A calculate node, that is, include a timestamp in the data update request that calculate node is sent, which sends for calculate node To the time of the first memory node transmission data update request, the first memory node is receiving the first calculate node and second When the data update for the first data that calculate node is sent is asked, asked determining to send according to the timestamp in respectively request Write permission is authorized the side for first sending request by the sequencing asked.

In addition, the first memory node and the second memory node compete the first memory node write permission can also use it is existing Other competition mechanisms in technology, the embodiment of the present invention are not described in detail.

In above-mentioned technical proposal, the calculate node of request is first sent by authorizing the write permission of the first memory node, no Concurrency conflict can only be avoided, additionally it is possible to improve the efficiency of distributed system.

Optionally, in the embodiment of the present invention, also ensure the data of each memory node using primary storage node locking mechanisms Consistency, specifically, the first calculate node to the first data carry out data update operation when, first according to the second subregion View determines at least two memory node of the first data of storage, wherein (also known as including a primary storage node：Cephalomere Point), the first calculate node first has to the write permission that competition obtains primary storage node, then could carry out data in primary storage node Update operation, after receiving the data update success message that primary storage node returns, then is located in chain structure shown in Fig. 2 In memory node after primary storage node carry out data update operation, every time the previous node only in chain structure (or Referred to as father node) write-in data success after, just in latter node carry out data update operation, to ensure data in more copies Consistency.

Optionally, in the embodiment of the present invention, also ensure the stability and validity of data using version control mechanism.Version This control refers to that the data being written every time correspond to a version number and first read host node when calculate node carries out write operation The version value V of middle data_i, the version value that data are written in this write operation is V_i+1.After introducing version control mechanism, storage section Point asks the covering of highest version data by lowest version data are not received, and ensures the stability and consistency of data.

Optionally, in the embodiment of the present invention, ensure the consistency of data and steady in conjunction with competition mechanism and Version Control It is qualitative.

Give storing three nodes of the first data in the first subregion view for example, might as well set as node A, section below Point B (being the first memory node), node C.Its interior joint A is main memory node, and in the first calculate node according to first point Before area view carries out data update, the versions of the first data for including in three nodes is t, and the states of the first data can be with It is labeled as：{A[t]、B[t]、C[t]}.

First calculate node is according to the request being updated to the first data at the first subregion view customer in response end, first The original version value for reading the first data in node A is t, and then, the request of the first calculate node is in node A by the first data update The data for being t+1 for version.After first calculate node carries out data update success in node A, continuing the transmission into node B will First data update is the request of the data of version t+1.

Before the data update for receiving the transmission of the first calculate node of node B is asked, with reference to Fig. 5, management node inspection It measures node A to break down, using node B as new primary storage node, and transient node (label is determined according to hash algorithm For hint) replacing node A, (hint nodes can use heartbeat detection mechanism by data copy to node A, and the embodiment of the present invention is to this Not be described in detail), and generate the second subregion view, three nodes that the first data are stored in the second subregion view be respectively node B, Hint nodes, node C.

After second calculate node receives the second subregion view of management node transmission, pair of client transmission is received The request that first data are updated determines that primary storage node is node B, at this point, section might as well be set according to the second subregion view Point B not yet receives the data update request of the first calculate node transmission, therefore, the corresponding version of the first data in node B T is remained as, therefore, the second calculate node is read in node B after the version of the first data, is sent to node B by the first data more The request of the new data for version t+1.

In this case, node B sends the request for receiving the transmission of the second calculate node and the first calculate node Request, the two request write-in data version it is legal, node B authorizes write permission the request first received corresponding meter Operator node.

If the second calculate node competes successfully, the second calculate node will be sequentially in node B, hint node and node It is updated the data using second in C and the first data is updated, the state of updated data is { B [t+1], hint [t+1], C [t+1]}.Since the version number of data in node B has been t+1, then version is written in the request for refusing the first calculate node by node B This is the request of t+1, avoids the covering of same version number data, and the first calculate node will be written to client returned data to fail.

Corresponding, if the first calculate node competes successfully, the first calculate node will be sequentially in node B and node C Middle updated the data using first is updated the first data, and state of the updated data in the first subregion view is { A [t +1]、B[t+1]、C[t+1]}.Since the version number of data in node B has been t+1, then node B will refuse the second calculate node Request write-in version be t+1 request, avoid the covering of same version number data, the second calculate node that from returning to number to client Fail according to write-in.

In above-mentioned technical proposal, by combining Version Control and locking mechanisms, it can either allow to regard based on the second subregion The calculate node of figure carries out data update operation when there is the task based on the first subregion view to be not finished, and can guarantee second point In area's view in different memory nodes data consistency.

Optionally, if the second calculate node compete the first memory node write permission failure, the second calculate node to Failed message is written in client returned data.

Optionally, as another embodiment, at least two storage sections of the first data of storage of the second subregion view instruction Point includes the first memory node and the second memory node；

Step 102：Second calculate node is updated the data using second according to the second subregion view and is carried out more to the first data New operation, when it is implemented, with reference to Fig. 6, includes the following steps：

Step 1023：Second calculate node is updated the data using second and is carried out to the first data that the first memory node stores Update forms the data of the first version in the first memory node；

Step 1024：Second calculate node determines that the corresponding version of the first data that the second memory node preserves is less than first The previous version of version；

Step 1025：Before the first data that second calculate node is preserved in the second memory node are updated to first version After the data of one version, using second update the data to the data of the previous version of the first version in the second memory node into Row update, forms the data of the first version in the second memory node.

Specifically, the second calculate node determines that at least two storages of the first data of storage save according to the second subregion view Point specifically includes the first memory node and the second memory node.Wherein, in second the first memory node of calculate node pair Before first data are updated, the first data correspond to the previous version of first version, and the second calculate node using second more After the first data in the first memory node of new data pair are updated, updated second, which updates the data corresponding version, is First version.

Second calculate node continues to be updated the first data in the second memory node, however, it was found that the second storage section The corresponding version of the first data in point is less than the previous version of first version, and in the embodiment of the present invention, the second calculate node will The data that the first data in the second memory node are updated to the previous version of first version are waited for then to reuse second It updates the data and the data of the previous version of the first version in the second memory node is updated.

In above-mentioned technical proposal, by forbidding the data cover of cross-version, i.e., only receive next version of original version value Data update is asked, and ensures the consistency of data, and when multiple calculate nodes carry out data update, avoid interfering other meters The successful execution of the data update of operator node.

It is asked although the second calculate node needs the first data of the second memory node of waiting to be updated to the second calculate node The last revision of the version of write-in is sought, still, the second calculate node can directly carry out data update in the first memory node, Without waiting for the first calculate node to execute the operation of the data update based on the first subregion view, remain able at reduction business Manage the time.

Optionally, in the embodiment of the present invention, step 1024：Second calculate node determines first that the second memory node preserves The corresponding version of data is less than the previous version of first version, when it is implemented, including the following steps：

Second calculate node is updated to the second memory node transmission data asks, and data update request is for asking with second It updates the data and is updated as the first data of data pair of first version；

Second calculate node receives the error information that the second memory node returns, and error information shows in the second memory node The corresponding version of the first data be less than first version previous version.

Step 1025：Before the first data that second calculate node is preserved in the second memory node are updated to first version After the data of one version, using second update the data to the data of the previous version of the first version in the second memory node into Row update, when it is implemented, including the following steps：

After second calculate node sets duration after receiving error information, updated again to the second memory node transmission data Request；

Second calculate node receives the data update success message that the second memory node returns, wherein the second memory node After the first data of storage are updated to the data of previous version of first version by the first calculate node, response second calculates Node send data update request, and the first data be updated to second update the data after to the first calculate node return number According to being updated successfully message.

Specifically, the second memory node will return after the data update request for receiving cross-version to the second calculate node Error information is returned, so that the second calculate node knows current data update request cross-version mistake.

The first data in second calculate node the second memory node of waiting are updated to the previous version of first version Data, realization method can be：

First, the second calculate node according to the setting period, is periodically updated to the second memory node retransmission data and is asked, After the version of the first data in the second memory node is updated to the data of previous version of first version, second calculates section The data update request that point is retransmitted will be accepted, and the second memory node is first version by version is updated in the first data After data message is updated successfully to the second calculate node returned data.

Second, it is first version that active is updated to version by the second memory node in the first data by the first calculate node After the data of previous version, the data that the first data have been updated to the previous version of first version are returned to the second calculate node Message, the second calculate node will be updated to the second memory node transmission data again asks, and the second memory node will accept this Request executes data update.

By above two mode, the second calculate node can be in time in the first data of the second memory node storage by more After the new data for the previous version of first version, operation is updated to the first data in the second memory node, reduces industry It is engaged in processing time.

In actual conditions, there is the situation of above-mentioned cross-version mistake in data update request, including：

Situation 1, in newly-increased memory node come when storing the first data, management module will be after instruction the first calculate node use The mode of platform copy, by the old version copy of the first data in newly-increased memory node, in this case, data update Request is likely to occur cross-version mistake, is illustrated below.

It might as well set and store the node of the first data in the first subregion view as node A (being the first memory node), node C, wherein node A is main memory node, and node C is tail node.It is distributed, is increased newly automatically according to user's request or management node Nodes of the node D as the first data of storage, wherein using node D as the child node of node A and the father node of node C.

With reference to Fig. 7, management node distributes the first calculate node and carries out background copy, the old version data of node C are copied Shellfish is to node D.It might as well set, before increasing node D newly, the state of data is { A [4], C [4] }, then the first calculate node is by first The data C [4] of data C [1] to the 4th version of a version be copied to node D (due to the consistency of copy, A [i]=C [i], I takes 1,2,3,4).

In addition, when the first calculate node carries out background copy, due to the data not yet one on node A, node D, node C It causes, management node will not generate the second subregion view, and calculate node carries out data processing still according to the first subregion view.It there is no harm in If third calculate node updates data, the state of data becomes { A when the first calculate node carries out background copy [5], [5] C | D [3] }, wherein D [3] shows that the third edition data C [3] of data is copied to by the first calculate node Node D.

After the first calculate node by the 5th of data the edition data C [5] by node D is copied to, the state of data becomes For { A [5], C [5] | D [5] }, after management node determines that the versions of data of node D is consistent with the versions of data of node C, the is generated Two subregion views, and the second subregion view is sent to all calculate nodes including the second calculate node.

But the time that different calculate nodes receive the second subregion view is variant, for example, in T₁Moment, the second meter Operator node receives the second subregion view；In T₁T after moment₂Moment, third calculate node not yet receive the second subregion and regard Figure, the update to data for but receiving client are asked, and third calculate node will be docked according to the first subregion view of preservation Data in point A and node C are updated.In T₂T after moment₃At the moment, third calculate node is according to the first subregion view It is { A [6], C [6] | D [5] } by data update.

T₃T after moment₄Moment, the second calculate node receive the request of client being updated to data, and second The version that calculate node reads the data of primary storage node A is 6, then sends the data of the version 7 of write-in data to node A, write It is { A [7], D [5], C [6] } according to the state of the second subregion viewdata after entering successfully.In T₄T after moment₅Moment, the Two calculate nodes continue to send the data of the version 7 of write-in data to node D, however, the version of data is 5 in node D, it will not The write request for receiving cross-version, to the second calculate node return error information, inform the second calculate node write operation exist across Version error.

In actual conditions, in T before₃Moment, third calculate node by data update be { A [6], C [6] | D [5] } it Afterwards, management node will assign the first calculate node to carry out background copy again according to the first subregion view, and C [6] is copied node D.And in T₅The background copy at moment, the first calculate node is not yet completed, thus just occur the second calculate node write operation across The mistake of version.Can periodically (version 7 of data be as written to the update request of node D retransmission datas in second calculate node Data request), in T₅T after moment₆At the moment, for the first calculate node successfully by C [6] copy nodes D, second calculates section The data update request of point will become legal, and continuation is carried out data update by the second calculate node in node D and node D, The state of data is { A [7], D [7], C [7] } after update.

In situation 1, using technical solution provided by the invention, the second calculate node can the first calculate node according to Before the data background copy that first subregion view carries out is completed, operation is updated to data according to the second subregion view, is subtracted The business stand-by period is lacked, has reduced business processing and take, improve system effectiveness.

Situation 2 deletes the situation of storage node.

It might as well set and store the node of data in the first subregion view as node A (being the first memory node), node B, section Point C (being the second memory node), according to user's instruction or management node Automatic dispatching, is removed from it node B, i.e., no longer will section Memory nodes of the point B as data, management node will generate the second subregion view, and the second subregion view is sent to each calculating Node.

But the time that different calculate nodes receive the second subregion view is variant, for example, in T₁Moment, the second meter Operator node receives the second subregion view；In T₁T after moment₂Moment, the first calculate node not yet receive the second subregion and regard Figure, the update to data for but receiving client are asked, and the first calculate node will be docked according to the first subregion view of preservation Data in point A and node C are updated.It might as well set, in T₂Before moment, the state of data is { A [2], B [2], C [2] }, then the data update of the first calculate node be written data version 3 data.

In T₃At the moment, data success is updated to A [3] by the first calculate node in node A, and data update is still in node C Not starting (or not yet completing), the update to data that the second calculate node receives client is asked, in the embodiment of the present invention, Second calculate node terminates without waiting for data update operation of first calculate node based on the first subregion view, is directly based upon the Two subregion views carry out data update operation.

First, the version that the second calculate node reads the data in primary storage node A is version 3, then asks to send out to node A The request for sending the data of the edition 4 of write-in data after being written successfully, continues in node C according to the second subregion view into line number According to update.However, when sending the request of the data of the edition 4 of write-in data to node C in the second calculate node, first calculates section Data update operation of the point in node C not yet terminates, and therefore, according to the second subregion view, the state of data is { A [4], C [2] }, there is cross-version mistake in the data update request of the second calculate node, and the second calculate node will wait for the first calculate node After being C [3] by the data update in node C, continue to be C [4] by data update in node C.

In situation 2, using technical solution provided by the invention, the second calculate node can the first calculate node according to Before the data update that first subregion view carries out is completed, operation is updated to data according to the second subregion view, is reduced The business stand-by period reduces business processing and takes, improves system effectiveness.

Based on the same technical idea, the embodiment of the present invention additionally provides data processing equipment in a kind of distributed system 200, distributed system includes management node, the first calculate node, data processing equipment 200 and multiple memory nodes, management section It is communicated between point, the first calculate node, data processing equipment 200 and multiple memory nodes, with reference to Fig. 8, data processing equipment 200 wraps It includes：

Receiving module 201, for being updated the data to first using first according to the first subregion view in the first calculate node When data are updated operation, the second subregion view is received；

Update module 202 is updated behaviour for being updated the data using second according to the second subregion view to the first data Make；

Optionally, in the embodiment of the present invention, at least two storage sections of the first data of storage of the first subregion view instruction Include the first memory node in point and at least two memory nodes of the first data of storage of the second subregion view instruction；

Update module 202, is specifically used for：

The write permission of the first memory node is competed with the first calculate node；

After the write permission for obtaining the first memory node, updated the data to the first number in the first memory node using second It is operated according to being updated.

Optionally, in the embodiment of the present invention, the first memory node is the first data of storage of the first subregion view instruction Memory node at least two memory nodes in addition to primary storage node, and the first memory node indicates for the second subregion view The first data of storage at least two memory nodes in primary storage node；

Update module 202 is specifically used for：

Determine that the primary storage node of the first data of storage is the first memory node according to the second subregion view；

The request that the first data are updated with operation is sent to the first memory node；

Optionally, in the embodiment of the present invention, at least two storage sections of the first data of storage of the second subregion view instruction Point includes the first memory node and the second memory node；

Update module 202 is specifically used for：

The first data for storing the first memory node are updated the data using second to be updated, and form the first memory node In first version data；

Determine that the corresponding version of the first data of the second memory node preservation is less than the previous version of first version；

After the first data that the second memory node preserves are updated to the data of previous version of first version, use Second updates the data and is updated to the data of the previous version of the first version in the second memory node, forms the second storage section The data of first version in point.

Optionally, in the embodiment of the present invention, update module 202 is specifically used for：

To the second memory node transmission data update ask, data update request for ask using second update the data as The first data of data pair of first version are updated；

The error information of the second memory node return is received, error information shows the first data pair in the second memory node The version answered is less than the previous version of first version；

After setting duration after receiving error information, updates ask to the second memory node transmission data again；

Receive the second memory node return data update success message, wherein the second memory node in storage first After data are updated to the data of previous version of first version by the first calculate node, the number of response data processing unit transmission Asked according to update, and the first data be updated to second update the data after be updated successfully and disappear to the first calculate node returned data Breath.

The data processing method corresponding with Fig. 3 of data processing equipment 200 in the present embodiment is to be based on same inventive concept Under two aspect, the implementation process of method is described in detail in front, so those skilled in the art can root The structure and implementation process of the data processing equipment 200 in the present embodiment are well understood according to foregoing description, in order to illustrate book Succinctly, details are not described herein again.

Based on identical inventive concept, the embodiment of the present invention also provides calculate node 300 in a kind of distributed system, application In distributed system, distributed system includes management node, the first calculate node, calculate node 300 and multiple memory nodes, It is communicated between management node, the first calculate node, calculate node 300 and multiple memory nodes, with reference to Fig. 9, calculate node 300 is wrapped It includes：Processor 301 and input/output interface 302.

Wherein, processor 301 are communicated with input/output interface 302, for being regarded according to the first subregion in the first calculate node Figure is updated the data using first when being updated operation to the first data, and instruction input/output interface 302 receives the second subregion and regards Figure；

Input/output interface 302 is used to be indicated to receive the second subregion view according to processor 301；

Input/output interface 302 is additionally operable to reception second and updates the data；

Processor 301 is additionally operable to be updated the data using second according to the second subregion view is updated behaviour to the first data Make；

Processor 301 updates the data using second according to the second subregion view and is updated operation to the first data, including：

Processor 301 and the first calculate node compete the write permission of the first memory node, including：

Indicate that input/output interface 302 sends the request that the first data are updated with operation to the first memory node；

Input/output interface 302 is additionally operable to send the request for being updated the first data operation to the first memory node；

Optionally, in the embodiment of the present invention, processor 301 determines the corresponding version of the first data that the second memory node preserves This is less than the previous version of first version, including：

It indicates that input/output interface 302 is updated to the second memory node transmission data to ask, data update is asked for asking The first data of data pair updated the data using second as first version are asked to be updated；

Indicate that input/output interface 302 receives the error information that the second memory node returns, error information shows that second deposits Store up the previous version that the corresponding version of the first data in node is less than first version；

Input/output interface 302 is additionally operable to update to the second memory node transmission data and ask；Receive the second storage The error information that node returns；

The first data that processor 301 is preserved in the second memory node are updated to the data of the previous version of first version Later, it is updated the data using second and the data of the previous version of the first version in second memory node is updated, including：

After setting duration after receiving error information, instruction input/output interface 302 is sent to the second memory node again Data update is asked；

Indicate that input/output interface 302 receives the data update success message that the second memory node returns；

Input/output interface 302 is additionally operable to update to the second memory node transmission data again and ask；Receive the second storage The data update success message that node returns；

Wherein, the second memory node is updated to by the first calculate node the previous version of first version in the first data of storage After this data, the data update request that the second calculate node of response is sent, and it is updated to the second update in the first data After data message is updated successfully to the first calculate node returned data.

It should be noted that the above processor 301 can be a processor, can also be the general designation of multiple processors. For example, processor 301 can be central processing unit, can also be specific integrated circuit, or be arranged to implement the present invention One or more integrated circuits of embodiment, such as：One or more microprocessors, or, one or more field-programmable Gate array.

Optionally, processor 301 is connect with input/output interface 302 by bus, which can be industrial standard body Architecture bus, external equipment interconnection bus or extended industry standard architecture bus etc..It is total that the bus can be divided into address Line, data/address bus, controlling bus etc..Only indicated with a line for ease of indicating, in figure, it is not intended that an only bus or A type of bus.

The data processing method corresponding with Fig. 3 of calculate node 300 in the present embodiment is based under same inventive concept Two aspects, are in front described in detail the implementation process of method, so those skilled in the art can be before The structure and implementation process of the calculate node 300 that description is well understood in the present embodiment are stated, in order to illustrate the succinct of book, herein Just repeat no more.

Based on identical inventive concept, the embodiment of the present invention provides a kind of distributed system, and distributed system includes management Node, the first calculate node, the second calculate node and multiple memory nodes, management node, the first calculate node, second calculate section It is communicated between point and multiple memory nodes；

Management node is used to indicate the first of at least two memory nodes that the first moment stored the first data for generating Subregion view, and generate and be used to indicate for the second moment and store the second subregions of at least two memory nodes of the first data and regard Figure, wherein the first moment is different from the second moment；First moment stored at least two memory nodes of the first data and when second At least two memory nodes for carving the first data of storage are not exactly the same；

First calculate node is updated behaviour for being updated the data using first according to the first subregion view to the first data Make；

Second calculate node, for being updated the data to first using first according to the first subregion view in the first calculate node When data are updated operation, the second subregion view is received, is updated the data using second according to the second subregion view and is counted to first It is operated according to being updated.

Second calculate node is used for：It is updated the data using second according to the second subregion view and behaviour is updated to the first data Make, including：

Distributed system data processing method corresponding with Fig. 3 in the present embodiment is based on two under same inventive concept A aspect is in front described in detail the implementation process of method, so those skilled in the art can be according to aforementioned The structure and implementation process of the distributed system in the present embodiment is well understood in description, in order to illustrate the succinct of book, herein It repeats no more.

It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, system or computer program Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention Apply the form of example.Moreover, the present invention can be used in one or more wherein include computer usable program code computer The computer program production implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) The form of product.

The present invention be with reference to according to the method for the embodiment of the present invention, the flow of equipment (system) and computer program product Figure and/or block diagram describe.It should be understood that can be realized by computer program instructions every first-class in flowchart and/or the block diagram The combination of flow and/or box in journey and/or box and flowchart and/or the block diagram.These computer programs can be provided Instruct the processor of all-purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce A raw machine so that the instruction executed by computer or the processor of other programmable data processing devices is generated for real The device for the function of being specified in present one flow of flow chart or one box of multiple flows and/or block diagram or multiple boxes.

These computer program instructions, which may also be stored in, can guide computer or other programmable data processing devices with spy Determine in the computer-readable memory that mode works so that instruction generation stored in the computer readable memory includes referring to Enable the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one box of block diagram or The function of being specified in multiple boxes.

Claims

1. data processing method in a kind of distributed system, which is characterized in that the distributed system includes management node, first Calculate node, the second calculate node and multiple memory nodes, the management node, first calculate node, second meter It is communicated between operator node and the multiple memory node, the method includes：

It is updated the data using first according to the first subregion view in first calculate node and operation is updated to the first data When, second calculate node receives the second subregion view；

Second calculate node is updated the data using second according to the second subregion view and is carried out more to first data New operation；

Wherein, the first subregion view and the second subregion view are generated by the management node；Described first point Area's view is used to indicate at least two memory nodes for storing first data at the first moment；The second subregion view is used for Indicate that the second moment stored at least two memory nodes of first data；Wherein, when first moment is with described second It carves different；First moment stores at least two memory nodes of first data and second moment storage described the At least two memory nodes of one data are not exactly the same.

2. the method as described in claim 1, which is characterized in that storage first data of the first subregion view instruction At least two memory nodes and the second subregion view instruction storage first data at least two memory nodes In include the first memory node；

Second calculate node is updated the data using second according to the second subregion view and is carried out more to first data New operation, including：

Second calculate node is updated the data using described second to institute after the write permission for obtaining first memory node First data stated in the first memory node are updated operation.

3. method as claimed in claim 2, which is characterized in that first memory node indicates for the first subregion view Storage first data at least two memory nodes in memory node in addition to primary storage node, and described first deposits Store up the primary storage section at least two memory nodes of storage first data that node is the second subregion view instruction Point；

Second calculate node determines that it is institute to store the primary storage node of first data according to the second subregion view State the first memory node；

Second calculate node sends the request that first data are updated with operation to first memory node；

Wherein, primary storage section of first calculate node in storage first data indicated the first subregion view First data in point are updated after operation, and first calculate node is sent to first memory node to institute State the request that the first data are updated operation；First memory node will be updated operation to first data and write Permission authorizes the corresponding calculate node of request for being updated operation to first data first received.

4. the method as described in claim 1, which is characterized in that storage first data of the second subregion view instruction At least two memory nodes include the first memory node and the second memory node；

Second calculate node updates the data first data stored to first memory node using described second It is updated, forms the data of the first version in first memory node；

It is described that second calculate node determines that the corresponding version of first data that second memory node preserves is less than The previous version of first version；

First data that second calculate node is preserved in second memory node are updated to the first version Previous version data after, updated the data to the first version in second memory node using described second The data of previous version are updated, and form the data of the first version in second memory node.

5. method as claimed in claim 4, which is characterized in that second calculate node determines that second memory node is protected The corresponding version of first data deposited is less than the previous version of the first version, including：

Second calculate node is updated to the second memory node transmission data asks, and the data update request is for asking The data updated the data using described second as first version are asked to be updated first data；

Second calculate node receives the error information that second memory node returns, and the error information shows described the The corresponding version of first data in two memory nodes is less than the previous version of the first version；

First data that second calculate node is preserved in second memory node are updated to the first version Previous version data after, updated the data to the first version in second memory node using described second The data of previous version are updated, including：

Second calculate node after setting duration, is sent to second memory node again after receiving the error information The data update request；

Second calculate node receives the data update success message that second memory node returns, wherein described second Memory node is updated to by first calculate node previous version of the first version in first data of storage After data, the data update request that second calculate node is sent is responded, and be updated in first data Described second update the data after to first calculate node return to the data update success message.

6. data processing equipment in a kind of distributed system, which is characterized in that the distributed system includes management node, first Calculate node, the data processing equipment and multiple memory nodes, the management node, first calculate node, the number It is communicated according between processing unit and the multiple memory node, the data processing equipment includes：

Receiving module, for being updated the data to the first data using first according to the first subregion view in first calculate node When being updated operation, the second subregion view is received；

Update module is updated behaviour for being updated the data using second according to the second subregion view to first data Make；

7. device as claimed in claim 6, which is characterized in that storage first data of the first subregion view instruction At least two memory nodes and the second subregion view instruction storage first data at least two memory nodes In include the first memory node；

The update module, is specifically used for：

After the write permission for obtaining first memory node, updated the data in first memory node using described second First data be updated operation.

8. device as claimed in claim 7, which is characterized in that first memory node indicates for the first subregion view Storage first data at least two memory nodes in memory node in addition to primary storage node, and described first deposits Store up the primary storage section at least two memory nodes of storage first data that node is the second subregion view instruction Point；

The update module is specifically used for：

Determine that the primary storage node for storing first data is first memory node according to the second subregion view；

9. device as claimed in claim 6, which is characterized in that storage first data of the second subregion view instruction At least two memory nodes include the first memory node and the second memory node；

The update module is specifically used for：

First data for storing first memory node are updated the data using described second to be updated, described in formation The data of first version in first memory node；

Determine that the corresponding version of first data of the second memory node preservation is less than the previous version of the first version This；

The data of the previous version of the first version are updated in first data that second memory node preserves Later, using described second update the data to the data of the previous version of the first version in second memory node into Row update, forms the data of the first version in second memory node.

10. device as claimed in claim 9, which is characterized in that the update module is specifically used for：

It updates and asks to the second memory node transmission data, the data update request is updated for asking with described second Data are updated first data as the data of first version；

The error information that second memory node returns is received, the error information shows the institute in second memory node State the previous version that the corresponding version of the first data is less than the first version；

After receiving the error information after setting duration, the data update is sent to second memory node again and asked It asks；

Receive the data update success message that second memory node returns, wherein second memory node is in storage After first data are updated to the data of previous version of the first version by first calculate node, described in response Data processing equipment send the data update request, and first data be updated to described second update the data after The data update success message is returned to first calculate node.

11. calculate node in a kind of distributed system, which is characterized in that be applied in distributed system, the distributed system Further include management node, the first calculate node and multiple memory nodes, it is the management node, first calculate node, described It is communicated between calculate node and the multiple memory node, the calculate node includes：

Processor is communicated with input/output interface, for using first according to the first subregion view in first calculate node It updates the data when being updated operation to the first data, indicates that the input/output interface receives the second subregion view；

The processor is additionally operable to be updated the data using second according to the second subregion view and be carried out more to first data New operation；

12. calculate node as claimed in claim 11, which is characterized in that the storage of the first subregion view instruction described the At least two of storage first data that at least two memory nodes of one data and the second subregion view indicate deposit Store up in node includes the first memory node；

The processor updates the data using second according to the second subregion view and is updated operation to first data, Including：

13. calculate node as claimed in claim 12, which is characterized in that first memory node is that first subregion regards Memory node at least two memory nodes of storage first data of figure instruction in addition to primary storage node, and it is described First memory node is the master at least two memory nodes of storage first data of the second subregion view instruction Memory node；

Indicate that the input/output interface is updated operation to first data to first memory node transmission and asks It asks；

The input/output interface is additionally operable to send to first memory node carries out the update behaviour to first data The request of work；

14. calculate node as claimed in claim 11, which is characterized in that the storage of the second subregion view instruction described the At least two memory nodes of one data include the first memory node and the second memory node；

15. calculate node as claimed in claim 14, it is characterised in that：

The processor determines that the corresponding version of first data that second memory node preserves is less than the first edition This previous version, including：

It indicates that the input/output interface is updated to the second memory node transmission data to ask, the data update request is used First data are updated using the data that described second updates the data as first version in request；

Indicate that the input/output interface receives the error information that second memory node returns, the error information shows institute State the previous version that the corresponding version of first data in the second memory node is less than the first version；

The input/output interface is additionally operable to send the data update request to second memory node；Receive institute State the error information of the second memory node return；

First data that the processor is preserved in second memory node are updated to the previous of the first version After the data of version, the previous version to the first version in second memory node is updated the data using described second This data are updated, including：

After receiving the error information after setting duration, indicate the input/output interface again to second memory node Send the data update request；

The input/output interface is additionally operable to send the data update request to second memory node again；Receive institute State the data update success message of the second memory node return；

Wherein, second memory node is updated to described first in first data of storage by first calculate node After the data of the previous version of version, data update request is responded, and be updated to described the in first data Two update the data after to first calculate node return to the data update success message.

16. a kind of distributed system, which is characterized in that the distributed system includes management node, the first calculate node, second Calculate node and multiple memory nodes, the management node, first calculate node, second calculate node and described more It is communicated between a memory node；

The management node is used to indicate the first of at least two memory nodes that the first moment stored the first data for generating Subregion view, and generate the second subregion for being used to indicate at least two memory nodes that the second moment stored first data View, wherein first moment is different from second moment；First moment stores at least the two of first data At least two memory nodes that a memory node stores first data with second moment are not exactly the same；

First calculate node carries out more first data for being updated the data using first according to the first subregion view New operation；

Second calculate node, for being updated the data pair using first according to the first subregion view in first calculate node When first data are updated operation, the second subregion view is received, the second update is used according to the second subregion view Data are updated operation to first data.

17. system as claimed in claim 16, which is characterized in that storage first number of the first subregion view instruction According at least two memory nodes and the second subregion view instruction storage first data at least two storage sections It include the first memory node in point；

Second calculate node is used for：According to the second subregion view using second update the data to first data into Row update operation, including：

18. system as claimed in claim 16, which is characterized in that storage first number of the second subregion view instruction According at least two memory nodes include the first memory node and the second memory node；